Signal-to-Noise Ratio, Variability, and Their Relevance in Clinical Trials
J Clin Psychiatry 2013;74(5):479-481
© Copyright 2015 Physicians Postgraduate Press, Inc.
Purchase This PDF for $40.00
If you are not a paid subscriber, you may purchase the PDF.
(You'll need the free Adobe Acrobat Reader.)
Receive immediate full-text access to JCP. You can subscribe to JCP online-only ($86) or print + online ($156 individual).
With your subscription, receive a free PDF collection of the NCDEU Festschrift articles. Hurry! This offer ends December 31, 2011.
If you are a paid subscriber to JCP and do not yet have a username and password, activate your subscription now.
As a paid subscriber who has activated your subscription, you have access to the HTML and PDF versions of this item.
Click here to login.
Did you forget your password?
Still can't log in? Contact the Circulation Department at 1-800-489-1001 x4 or send email
Some curious observations emerge from a scrutiny of clinical trial research in psychiatry. For example, small studies often yield statistically significant outcomes, whereas large studies often fail to separate experimental drug from placebo, there are plenty of failed trials even for drugs that are subsequently approved by regulatory authorities, and number needed to treat (NNT) values are often large, implying that medications play only a small role in eliciting clinical benefits. There are many good explanations for these 3 observations, and 1 involves the variability of the outcome measures in clinical trials. Such variability can be true or spurious. If the true value of the outcome measure is considered a signal, spurious variability in the signal is due to a phenomenon termed noise. There is little discussion on signal and noise in psychiatric literature. This article therefore discusses the signal-to-noise ratio, true and spurious variability, and their impact on clinical trial research as well as everyday clinical practice. When variability of the outcome measure is low, even small sample studies may yield statistically significant outcomes. When studies are conducted at many sites, variability due to noise increases, the value of the signal-to-noise ratio falls, and statistically significant outcomes may not be detected or, if they are detected, the NNT is inflated.